Phooom said:
I have a 2 newbie questions.
When judging a computer, I sometime looked at the GHz to see if it's
a good computer (i.e. to see if it's over 1 GHz or if it's at 2.5
GHz). However, I remember reading somewhere that GHz isn't a reliable
way to judge a computer.
What other factors I should look for when judging a computer? The only
other two factors I look at is really the amount of RAM and the type of
video card it has. Is there anything else I should take in
consideration?
And I'm slightly confused for AMD processor's performance. For example,
if it says 3800+ 2.0Ghz, does that mean it runs as efficient as 3.8 GHz
but really at 2.0 GHz?
Thanks!
As Kony explained, the clock rate and the IPC are the two factors
that affect performance. IPC (instructions per clock) is a measure
of parallelism and efficiency. There are multiple functional units
in both Intel and AMD processors. The AMD processor seems to make
better use of what it has got, than the Intel one, and that better
use, helps compensate for the lower clock rate of the AMD processor.
(I wish I knew enough about architecture to explain why that is.)
Intel's Pentium-M shares this characteristic with the AMD processors,
and the Pentium-M behaves more like an AMD processor, than the
rest of the P4 Netburst processors.
This will change a bit when Intel ships Conroe, and Conroe has
slightly better internal parallelism.
Adding functional units is not an arbitrary matter, as both
companies could flood the chip with more functional units if
they wished. The compiler has to make good use of them, and
it is pretty difficult to do better than what is presently
achieved. If additional functional units were added, they would
be idle most of the time. This is what makes Conroe a puzzle
to me.
The best way I can think of to judge the processors, is with a
benchmark that does the same kind of work as you plan to do.
Now, that is a hard thing, because many benchmarks are synthetic
or favor gaming. And other applications may depend on the type
of disk drives used, and not see nearly as much difference caused
by the CPU. (So if the computer has a 10000 RPM disk, that may
perform better than the average 7200 RPM disk.)
In addition to (clock_rate * IPC), there is also the issue of
how many processors are sitting on the silicon die. There are
single core and dual core processors. Many older software
programs would not run any faster in clock time, if they are
run on a dual processor. Photoshop can take advantage of two
processors, since an image can be chopped into two pieces, and
the two processors work on their piece independently. But not
enough programs yet make use of two processors, to make dual
core processors a consistent win.
Dual core processors also allow one "grinding task" to run on one
processor, while the other processor is available for you to
use for email, web surfing and so on. The computer is "more
responsive", since it can handle two major tasks before it
bogs down. That is "task level parallelism", and is good for
power users who can start many jobs running on the computer
at the same time. If the human operator is a "one thing at
a time" kind of person, then a dual core processor will not
do as much useful work for such a person.
These two charts demonstrate a computer program that can use
two processors, versus a program that currently only uses
one processor. Notice how the dual core processors are winning
in the first chart, while the single core processors are
winning in the second chart.
3D Studio Max - 3D modelling - uses the two processors in a dual
core processor, so the dual processors are at the top of the
chart:
http://www23.tomshardware.com/cpu.html?modelx=33&model1=238&chart=74&model2=212
http://www.3dtotal.com/team/tutorials/hugechair/chair1_01.asp
Winrar - file compression - uses one processor, so the fastest
processor core wins:
http://www23.tomshardware.com/cpu.html?modelx=33&model1=238&chart=62&model2=212
Dual core processors further complicate the selection process,
as you can see from the charts. My usage pattern, does not
include the purchase of a lot of new software, and I have many
programs I continue to use, that I bought years ago. A dual core
will not help me too much. Only you know what kind of programs
you are using or plan to use, and that may influence which kind
of processor you buy.
The video card selection is important for gaming, but for
2D office applications, the selection hardly matters. Some
cheap prebuilt computers, will use graphics contained in the
Northbridge chip, which is fine for office applications or
web surfing, but not for gaming. Note: If you buy a computer
with integrated (Northbridge) graphics and no video card,
make sure the computer has a video card slot! A favorite ploy
on cheap computers, is the absence of a video card slot, and
we have to disappoint one USENET poster after another, with the
bad news, that their new computer is not expandable when it
comes to adding a video card. Always check for a video card
slot on a prebuilt computer.
RAM still influences performance a bit, but wasting money on
"uber" RAM is hardly cost effective. Dual channel RAM (meaning
two separate hardware pathways from the memory sticks, to the
rest of the hardware) can supply additional memory bandwidth,
and might help with applications like Photoshop. For many other
applications, the effect is less pronounced. One difference
between them, is the maximum amount of RAM that can be populated.
Dual channel memory motherboards usually have room for 4 DIMMs,
which is better than the 3 DIMM slots on a single channel
motherboard.
Notice that, for the AMD "P.R." rating scheme (that 3800+
thing), that AMD gives itself an added boost to its number,
on its dual core processors. That assumes the user of the
computer, is using a program like Photoshop, that can harvest
all the additional performance of the dual core processors.
If all you did, was run Winrar all day, then the 3800+ number
would not be representative of the performance of the processor.
So, not only is the "P.R." number important, but the number
of cores is also important, especially if your usage pattern
will not make the most of dual cores. If you were a Photoshop
user, or 3D Studio Max, then the P.R. number is more meaningful
to you, as the 3800+ rating is the amount of speedup you would
see while using Photoshop.
If you want to "derate" an AMD dual core processor, to single
core usage pattern, try this:
This is an AMD X2 dual core, 3800+, 2GHz clock, 512KB L2 cache per core
http://www.amdcompare.com/us-en/desktop/details.aspx?opn=ADA3800DAA5BV
The nearest single core processor, with 2GHz clock, 512KB L2,
is this one. The P.R. rating here is 3200+
http://www.amdcompare.com/us-en/desktop/details.aspx?opn=ADA3200DAA4BP
Notice that I was careful to select socket 939 for both processors,
so the memory interface differences would not affect that rating.
What my above comparison means, is if you bought an X2 3800+ and
used only one of the cores (with Winrar say), then the processor
behaves closer to a single core Intel 3.2GHz processor.
So, by using the
www.amdcompare.com web site, you can get statistics
that will allow you to compare equivalent performance, depending
on your usage pattern. Or just use the Tomshardware charts, and get
real numbers that way. The Tomshardware chart avoids the need for the
customer, to do a lot of complicated comparisons, to pick a winner.
HTH,
Paul
